René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Binary file change by hook not detected due to Git textconv","articleBody":"### search you tried in the issue tracker\n\ntextconv\n\n### describe your issue\n\nI’ve set up Git to use `exiftool` as a text conversion filter for JPEGs, with this in `~/.config/git/config`:\r\n\r\n```ini\r\n[diff \"exiftool\"]\r\n    textconv = exiftool --composite -x 'Exiftool:*' -x 'File:*' -g0\r\n    cachetextconv = true\r\n    xfuncname = \"^-.*$\"\r\n```\r\n\r\n…and this in `~/.config/git/attributes`:\r\n\r\n```\r\n*.jpeg diff=exiftool\r\n```\r\n\r\nIn my repo, I’m using `jpegoptim` to optimize JPEG's with the below `.pre-commit-config.yaml`.\r\n\r\nTake an unoptimized JPEG:\r\n\r\n```console\r\n$ git status\r\nOn branch main\r\nChanges to be committed:\r\n        new file:   donut.jpeg\r\n```\r\n\r\nRunning pre-commit, the hook passes:\r\n\r\n```console\r\n$ pre-commit run jpegoptim\r\nOptimize JPEGs...........................................................Passed\r\n```\r\n\r\n(There’s no option to make `jpegoptim` exit with a non-zero status code when it changes files.)\r\n\r\nHowever, the tool did actually modify the image:\r\n\r\n```console\r\n$ git status\r\nOn branch main\r\nChanges to be committed:\r\n        new file:   donut.jpeg\r\n\r\nChanges not staged for commit:\r\n        modified:   donut.jpeg\r\n```\r\n\r\nThis behaviour means commits don't include the optimized version of a flie, instead they’re left around unstaged.\r\n\r\npre-commit doesn't detect the change because the text conversions of the binary file before/after are the same, leading to an empty diff:\r\n\r\n```\r\n$ git diff donut.jpeg\r\n```\r\n\r\nBut using [`--no-textconv`](https://git-scm.com/docs/git-diff#Documentation/git-diff.txt---no-textconv) shows the tool actually made a change:\r\n\r\n$ git diff --no-textconv donut.jpeg |cat\r\ndiff --git a/donut.jpeg b/donut.jpeg\r\nindex 3bd0743..d66a2f4 100644\r\nBinary files a/donut.jpeg and b/donut.jpeg differ\r\n```\r\n\r\nThe `git diff` call that pre-commit uses to check for changes, in [`_git_diff()`](https://github.com/pre-commit/pre-commit/blob/e846829992a84ce8066e6513a72a357709eec56c/pre_commit/commands/run.py#L273C6-L277) already uses `--no-ext-diff` to make diffs reproducible\r\nI think it should also use `--no-textconv`.\r\nThis would also make it a little faster.\n\n### pre-commit --version\n\n3.0.3 / main\n\n### .pre-commit-config.yaml\n\n```yaml\nrepos:\r\n-   repo: local\r\n    hooks:\r\n    -   id: jpegoptim\r\n        name: Optimize JPEGs\r\n        entry: jpegoptim\r\n        language: system\r\n        types: [jpeg]\n```\n\n\n### ~/.cache/pre-commit/pre-commit.log (if present)\n\n_No response_","author":{"url":"https://github.com/adamchainz","@type":"Person","name":"adamchainz"},"datePublished":"2023-02-02T11:00:55.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":0},"url":"https://github.com/2742/pre-commit/issues/2742"}